Research · The Decoder ·
New benchmark exposes how badly AI struggles with real knowledge work
A new benchmark finds that even the best AI model can fully solve only 3% of realistic knowledge-work tasks. The result highlights major gaps in current systems’ ability to handle practical office work reliably.